The themes
#8am-ai#index#themes#reference
David Olsson
A theme is a cluster of ideas the group kept returning to across meetings. Fourteen of them hold the corpus. Each is defined below with its size, span, and a few ideas cited straight from the record. The theme engagement page charts how each rose and fell over the two years.
agentic workflow
How agents got built, pointed, and let loose β the corpusβs clearest arc.
143 ideas Β· 2024-04-17 to 2026-06-17 Β· chapter: from scrubs to swarms
Cited:
- Shifting Focus to Cognitive Effort and Value in Work β 2026-04-29 (grade 70.8). David noted that Juan is addressing the quality problem by spending time in one-on-one sessions to work through the cognitive effort that went into producing assets.
- Difficulty in Collaborative Feedback Loops β 2026-04-29 (grade 70.8). David described the difficulty of working with feedback in the current LLM workflow, where criticisms like "this needs to change" don't carry enough context.
- Skills and Agents to Structure Development β 2026-04-08 (grade 69+). The group works through using skills and agents to put structure around otherwise unbounded AI development.
context engineering
The model is the commodity; the context is the lever.
97 ideas Β· 2024-04-24 to 2026-06-24 Β· chapter: the model is the commodity
Cited:
- Onboarding and UI challenges β 2026-06-10 (grade 75.2). David is using Project State for the "Stone Maps" project and names onboarding and the current interface as the open challenges.
- Bypassing Safeguards via Conditioning β 2026-05-27 (grade 74.5). Fulvio describes bypassing LLM safety filters during penetration testing by presenting assertions for the model to evaluate rather than asking it to act.
- Business Writing Strategy β 2026-06-24 (grade 72.9). Ying advises prioritizing a clear understanding of the target audience and defining the unique value proposition before writing.
product and business
From product to process; market access as the scarce thing.
61 ideas Β· 2024-06-05 to 2026-06-10 Β· chapter: product to process
Cited:
- Claude Versus OpenAI Utility β 2026-05-13 (grade 74.9). David explains preferring Claude as the product for knowledge workers and OpenAI as the standard for raw processing.
- Concerns about Bias and LLM Reliability in Business β 2026-03-11 (grade 69.3). In business, relying solely on an LLM is hard; a 90% success rate is not enough when predictability is the requirement.
- Strategic State and Customer Profiles β 2026-05-06. The structured-substrate idea applied to the business itself β a living model of customers and pipeline.
developer productivity
A working developerβs log of tool churn and the moving bottleneck.
59 ideas Β· 2024-08-28 to 2026-06-10 Β· chapter: the tool churn log
Cited:
- Career referral and development philosophies β 2026-06-10 (grade 75.7). Juan and David debate "vibe coding" versus formal development.
- Leveraging AI for Legacy Codebases β 2026-04-15 (grade 71.9). Building scaffolding around a mature codebase, leaning on Claude for production code.
- Utilizing Claude Code for Application Development β 2026-04-15 (grade 71.9). A rigorous spec written with Opus, then brought into Claude Code to build.
system vs design thinking
Atomic tasks to architecture β the spine of the corpus.
59 ideas Β· 2024-07-31 to 2026-06-24 Β· chapter: atomic to architecture
Cited:
- Operationalizing Meeting Insights β 2026-06-24 (grade 67.2). Moving from passive conversation to an automated repository that parses and stores transcripts.
- Shifting Focus to Problem-Solving β 2026-06-03 (grade 65.8). Juan argues technical work should shift from the language to identifying the right problem.
- Skills Development and Systemic AI Integration β 2025-12-17 (grade 65.6). Jason raises "Claude skills" β building systems to solve the problem of the problem.
ai and education
If the tools can do the work, what is school for. The groupβs conscience.
55 ideas Β· 2024-06-26 to 2026-06-24 Β· chapter: teaching in the rubble
Cited:
- Team Evolution and Learning Gaps β 2026-04-29 (grade 73.3). A learning loop where team members focus only on their piece and miss the whole.
- AI as a Trajectory Refiner β 2026-06-24 (grade 70.2). James frames AI as a trajectory refiner for motivated learners; Ying counters on foundational knowledge.
- Concerns about Learning and Skill Gaps β 2026-04-08. The false confidence of producing output without the knowledge to evaluate it.
meta / meeting cadence
The group examining its own cadence; the meeting as the artifact.
54 ideas Β· 2024-05-15 to 2026-06-24 Β· chapter: the meeting as method
Cited:
- Welcome and Meeting Purpose β 2026-06-24 (grade 69.4). David welcomes James back, the person who started the weekly meetings.
- Collective Memory and Future Value β 2026-06-17 (grade 67.2). Making an executable product from the meetings themselves.
- Introduction of the AI Group Collaboration Readiness Assessment Survey β 2026-03-11 (grade 65.9). Deploying a survey to the group from inside a chatbot.
signal harvesting
Reading across everything said to surface the glue.
40 ideas Β· 2024-08-14 to 2026-06-24 Β· chapter: mining the glue
Cited:
- Reflections on AI development β 2026-06-10 (grade 78.8). Creating artifacts and knowledge graphs is valuable for long-term intelligence even when LLMs hallucinate.
- Data ingestion and connectors β 2026-06-10 (grade 78.3). Project State integrates Slack, Google, and GitHub to gather project information.
- Project State's Four Layers and Signal Harvesting β 2026-04-22 (grade 74.4). Agents, schema, data, and time β the structure that makes harvesting repeatable.
people and team
The social layer where individual AI gains are won or lost.
24 ideas Β· 2024-08-14 to 2026-06-24 Β· chapter: the social layer
Cited:
- Role of LLMs in Professional Writing β 2026-06-24 (grade 75.3). James bridges investment and client teams using Claude as a drafting partner.
- Peer group software migration β 2026-06-10 (grade 72.6). Migrating the 8am meeting to software that automates email and builds a team knowledge graph.
- Remote Work and Tool Efficiency β 2026-05-20 (grade 68.6). The hidden cost of remote tool use and working through sickness unnoticed.
future and society
The philosophical edge β Neuralink, Ozempic, UBI, the human after.
15 ideas Β· 2024-08-07 to 2026-06-03 Β· chapter: the philosophical edge
Cited:
- Societal Implications of Human-AI Intersection β 2026-05-27 (grade 79.0). Neuralink, software-versus-hardware alignment, and the union of human and LLM thought-space.
- Human Skills and the Humanities β 2026-06-03 (grade 74.1). As the world goes digital, human-to-human connection becomes more valuable, not less.
- Historical Analogies for Societal Disruption β 2025-12-03 (grade 62.1). Fulvio compares AI's disruption to the Black Plague and the Industrial Revolution.
llm evaluation and trust
They donβt know how they know β so your input is the eval.
14 ideas Β· 2024-08-28 to 2026-06-24 Β· realized as experiment EXP-0001
Cited:
- Git Gists as Data Substrate β 2026-06-17 (grade 79.6). Using gists to hash and verify data β built into a runnable experiment.
- Evolution of Skill Requirements β 2026-06-24 (grade 72.6). "Prompt engineering" as a distinct skill is becoming a historical concept.
- LLM Self-Reflection and Human Evaluation β 2026-05-27 (grade 70.2). LLMs lack certainty about how they reach conclusions; human input is the eval.
hiring and authentication
Authenticating a person who runs an LLM during the interview.
12 ideas Β· 2024-07-31 to 2026-06-17 Β· realized as experiment EXP-0003
Cited:
- Blade Runner 2049 and Authentication Challenges β 2026-05-27 (grade 73.7). The baseline test as a frame for hiring when candidates can run an LLM live.
- Remote Interview Challenges β 2026-05-20 (grade 72.7). Candidates narrate convincingly; the interview stops testing the person.
- Impact of AI on Hiring and Credibility β 2026-06-03 (grade 71.8). Verifying candidates in a market saturated with AI tools.
project-state substrate
The structured-memory substrate the facility is built on.
8 ideas Β· 2025-06-25 to 2026-06-10 Β· chapter: mining the glue
Cited:
- Introduction of Project State β 2026-06-10 (grade 83.8). An application that delivers skills and schemas to manage projects and stakeholders.
- The Nature of LLMs as Artificial Simulants β 2026-05-27 (grade 75.6). LLMs exist on a provided manifest, not lived experience.
- Leveraging Digital Tracing and AI Models β 2025-10-22 (grade 69.0). Digital tracing of conversations for introspection and multi-vector analysis.
tool stack
The churn of named tools β what got tried and dropped.
8 ideas Β· 2024-09-11 to 2025-10-22 Β· covered in the tool churn log
Cited:
- AI Tools Integration & Implementation β 2025-09-17 (grade 61.7). Fireflies MCP integrated with Claude Desktop for transcript analysis.
- Challenges with AI Integration and Behavior Change β 2025-10-22 (grade 61.0). Integrating tools without increasing user workload.
- Leveraging Fireflies for Transcript Summaries β 2025-10-08 (grade 57.9). Generating summaries from ~800 transcripts.
The full per-theme synthesis lives in the repository at themes/THEME-NN/synthesis.md. Every idea cited here points back to the meeting it came from. The ideas index lists all 759 in full.